Trajectories collected on real or simulated robots.
Scalable synthetic video generation with rich ground truth.
15 million video frames from 7 robot platforms for large-scale visual learning.
